NoteTkr mirror of aufklarer/Pyannote-Segmentation-MLX This repository is a copy of https://huggingface.co/aufklarer/Pyannote-Segmentation-MLX/tree/abef0110277063f0ea117a802832a3eba22af84c made on 2026-11-12. The NoteTkr iOS app downloads these files from the original repository and turns to this copy only when the original repository refuses the download. The app checks every downloaded file against a SHA-256 fingerprint before use. Every file here except NOTICE.txt and LICENSE is byte-for-byte identical to that revision of the original repository. Nothing was modified. README.md is the original model card, unchanged, and describes the original repository. Model: pyannote/segmentation-3.0 (pyannote.audio). Copyright (c) 2023 CNRS. https://huggingface.co/pyannote/segmentation-3.0 Conversion to MLX: aufklarer (aufklarer/Pyannote-Segmentation-MLX). The conversion pre-computes the SincNet band-pass filters from the learned parameters, transposes convolution weights to channels-last, splits the bidirectional LSTM into forward and backward stacks with its two bias vectors summed, and renames the weight keys. License: MIT. The text is in LICENSE, reproduced from https://huggingface.co/pyannote/segmentation-3.0/blob/e66f3d3b9eb0873085418a7b813d3b369bf160bb/LICENSE The files are provided "as is", without warranties of any kind. The creators of the model and the publishers of the original repository do not endorse NoteTkr or this copy.